kandy-skills

Open-source Claude Code skills for KANDy — Kolmogorov-Arnold Networks for Dynamics, a scientific Python library for data-driven identification of dynamical systems (x_dot = A · Ψ(φ(x))).

These skills teach coding agents how to use KANDy correctly: choosing Koopman lifts, training single-layer KANs, extracting symbolic governing equations, and generating finite-volume PDE training data — with runnable examples for a dozen benchmark systems.

Skills

Skill	Description
kandy	System identification / equation discovery with KANDy: lift selection, training, symbolic extraction, FV numerics, and example scripts

Installation

As a plugin (recommended)

This repo is a Claude Code plugin and its own marketplace. In Claude Code:

/plugin marketplace add Center-For-Complex-Systems-Science/kandy-skills
/plugin install kandy@kandy-skills

Or from the terminal:

claude plugin marketplace add Center-For-Complex-Systems-Science/kandy-skills
claude plugin install kandy@kandy-skills

Manual copy

Alternatively, copy the skill directory into a skills folder:

# Project-level (shared with collaborators via git)
mkdir -p .claude/skills
cp -r skills/kandy .claude/skills/

# Or user-level (available in all your projects)
mkdir -p ~/.claude/skills
cp -r skills/kandy ~/.claude/skills/

Claude Code discovers the skill automatically from its SKILL.md frontmatter and loads the reference docs and examples on demand.

Layout

.claude-plugin/
├── plugin.json           # Claude Code plugin manifest
└── marketplace.json      # lets the repo be added as a plugin marketplace
skills/kandy/
├── SKILL.md              # entry point: core workflow + when to use what
├── references/
│   ├── lifts.md          # Koopman lift selection guide
│   ├── training.md       # optimizers, rollout loss, discrete maps, periodic phases
│   ├── symbolic.md       # symbolic extraction, custom libraries, scoring, LaTeX
│   ├── numerics.md       # finite-volume PDE data generation
│   └── api.md            # full public API reference
└── examples/             # complete, runnable scripts, grouped by fit recipe
    ├── odes/             # continuous ODEs (Lorenz, …)
    ├── maps/             # discrete maps (Hénon, Ikeda)
    ├── oscillators/      # coupled phase oscillators (Kuramoto, …)
    ├── pdes/             # 1D PDEs (Burgers, Kuramoto–Sivashinsky)
    ├── fluids/           # fluid dynamics (3D Navier–Stokes)
    ├── mathbio/          # mathematical biology (predator-prey)
    └── geometry/         # engineered lifts on manifolds (Hopf, trefoil)

Requirements

The skill assumes the kandy package is installed in the working environment:

pip install kandy

Python 3.11–3.13 · PyTorch ≥ 2.0 · PyKAN ≥ 0.2.0 · SciPy ≥ 1.10 · SymPy ≥ 1.12

License

MIT — see LICENSE.